Skip to content

RAG Knowledge Base Systems

A curated list of RAG (Retrieval-Augmented Generation) frameworks, platforms, and vector databases. Last updated: 2025.

RAG Frameworks

Developer-facing libraries for building RAG pipelines.

ProjectCompanyStarsLicenseDescription
LangChainLangChain Inc. (US, a16z)90k+MITDe facto standard LLM app framework. Modular chain architecture for RAG; LangGraph for agentic RAG, LangSmith for observability.
LlamaIndexLlamaIndex Inc. (US)38k+MITData framework purpose-built for RAG. 100+ connectors, hierarchical retrieval, multimodal RAG. Best for document Q&A.
Haystackdeepset (Germany)17k+Apache 2.0Enterprise end-to-end LLM/RAG pipeline framework. v2 redesigned in 2024; built-in evaluation, production-ready.

All-in-One Knowledge Base Platforms

ProjectCompanyStarsLicenseDescription
DifyLangGenius (China)45k+Apache 2.0Visual drag-and-drop LLM app platform with built-in RAG pipeline and knowledge base management. Supports 100+ models.
RAGFlowInfiniFlow (China)30k+Apache 2.0RAG engine based on deep document understanding. Excels at complex PDFs, tables, and charts. Supports citation tracing.
FastGPTlabring (China)20k+Apache 2.0LLM knowledge base platform with visual Flow workflow orchestration. TypeScript/Next.js, team collaboration support.
AnythingLLMMintplex Labs (US)30k+MITFull-stack desktop/server LLM app. Turns any document into a RAG chatbot. Multi-user, local+cloud LLM, multi-vector-DB.
MaxKBFIT2CLOUD (China)Apache 2.0Enterprise RAG Q&A bot supporting multiple LLM backends. From the 1Panel/JumpServer ecosystem.
QuivrQuivr HQ (France)36k+Apache 2.0"AI second brain" personal knowledge assistant. Multi-format RAG retrieval, cloud and local deployment.
PrivateGPTZylon (privacy AI)54k+Apache 2.0100% local, privacy-first document Q&A. Fully offline. One of the earliest viral local RAG projects.
KotaemonCinnamon AI (Vietnam/Japan)18k+Apache 2.0Clean, customizable document RAG chat tool with Gradio UI. Multi-vector retrieval, citation display, GraphRAG integration.
Onyx (Danswer)Onyx (YC W23)13k+MITConnects 50+ sources (Confluence/Notion/Slack). Self-hosted RAG enterprise Q&A platform, any LLM.
WeKnoraTencent (China)Apache 2.0Enterprise knowledge platform open-sourced by Tencent. "Wiki Mode" productizes Karpathy's LLM-as-Compiler pattern. Hybrid retrieval (vector + BM25 + knowledge graph), multi-tenant RBAC, built-in Agent, IM integration.

Vector Databases

ProjectCompanyStarsLicenseDescription
ChromaChroma (US, YC)16k+Apache 2.0Lightweight embedded vector database. Easiest to get started with for RAG prototyping.
WeaviateWeaviate B.V. (Netherlands, $50M+)11k+BSDVector database with GraphQL API. Supports hybrid search and multimodal retrieval.
QdrantQdrant (Germany, $28M Series A)21k+Apache 2.0High-performance vector search engine written in Rust. Excellent filtering for production RAG.
MilvusZilliz (US/China, $113M+)30k+Apache 2.0Cloud-native vector database for billion-scale vectors. Commercial: Zilliz Cloud.
pgvectorCommunity (Supabase contributor)PostgreSQLPostgreSQL vector extension with HNSW/IVFFlat index. Zero-cost vector retrieval if already on Postgres.

Managed Vector Database Services

ProductCompanyDescription
PineconePinecone Systems (US, $100M+)Fully managed serverless vector database. Zero ops, most mature managed RAG storage option. Closed source.
Zilliz CloudZilliz (Milvus parent)Managed cloud service for Milvus. Enterprise SLA and security.

Graph RAG

ProjectCompanyStarsDescription
GraphRAGMicrosoft Research19k+Graph-based RAG via community summaries. Answers complex multi-hop questions.
LightRAGHKU Data Intelligence Lab15k+Dual graph-vector index. Supports local/global dual-mode retrieval. Emerged late 2024.
UnWeaverCommunityEntity-level decomposition + aggregation on top of VectorRAG. Beats GraphRAG in factual correctness without building an explicit knowledge graph.
TrustGraphCommunityKnowledge graph + RAG platform. Combines structured graph reasoning with vector retrieval.

Vector-Free RAG

Alternative retrieval approaches that don't rely on embedding similarity.

ProjectCompanyStarsDescription
PageIndexVectifyAI29k+Semantic summary tree replaces vector database. Retrieval = reasoning via LLM tree search. 98.7% accuracy on FinanceBench.
MemvidCommunitySingle .mv2 file replaces the entire RAG pipeline and vector database. Sub-millisecond retrieval.
OhMyGoCommunityGo-based AI knowledge platform integrating chat, RAG Q&A, image recognition, and speech synthesis.
he-wiki-ragCommunityChapter-tree indexing preserves Markdown H1/H2/H3 hierarchy to avoid semantic loss from fixed-length chunking. Vector + BM25 + Cross-Encoder Rerank pipeline.

Traditional Search Engines (RAG-extended)

ProductCompanyDescription
ElasticsearchElastic N.V. (NYSE: ESTC)Enterprise search extended with ESRE and ELSER sparse vectors. Mature hybrid search (BM25 + vector) at billion-scale.
OpenSearchAWS (Amazon)AWS open-source Elasticsearch fork. Deep Amazon Bedrock integration. Neural Sparse Retrieval for cloud-native RAG.

Released under the MIT License.